[PATCH 15/18] ARC: mm: support 3 levels of page tables
Mike Rapoport
rppt at kernel.org
Wed Aug 11 05:24:36 PDT 2021
On Tue, Aug 10, 2021 at 05:42:55PM -0700, Vineet Gupta wrote:
> ARCv2 MMU is software walked and Linux implements 2 levels of paging: pgd/pte.
> Forthcoming hw will have multiple levels, so this change preps mm code
> for same. It is also fun to try multi levels even on soft-walked code to
> ensure generic mm code is robust to handle.
>
> overview
> ________
>
> 2 levels {pgd, pte} : pmd is folded but pmd_* macros are valid and operate on pgd
> 3 levels {pgd, pmd, pte}:
> - pud is folded and pud_* macros point to pgd
> - pmd_* macros operate on actual pmd
>
> code changes
> ____________
>
> 1. #include <asm-generic/pgtable-nopud.h>
>
> 2. Define CONFIG_PGTABLE_LEVELS 3
>
> 3a. Define PMD_SHIFT, PMD_SIZE, PMD_MASK, pmd_t
> 3b. Define pmd_val() which actually deals with pmd
> (pmd_offset(), pmd_index() are provided by generic code)
> 3c. Define pmd_alloc_one() and pmd_free() to allocate pmd
> (pmd_populate/pmd_free already exist)
>
> 4. Define pud_none(), pud_bad() macros based on generic pud_val() which
> internally pertains to pgd now.
> 4b. define pud_populate() to just setup pgd
>
> Signed-off-by: Vineet Gupta <vgupta at kernel.org>
> ---
> arch/arc/Kconfig | 4 ++
> arch/arc/include/asm/page.h | 11 +++++
> arch/arc/include/asm/pgalloc.h | 22 ++++++++++
> arch/arc/include/asm/pgtable-levels.h | 63 ++++++++++++++++++++++++---
> arch/arc/include/asm/processor.h | 2 +-
> arch/arc/mm/fault.c | 4 ++
> arch/arc/mm/tlb.c | 4 +-
> arch/arc/mm/tlbex.S | 9 ++++
> 8 files changed, 111 insertions(+), 8 deletions(-)
>
> diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig
> index 59d5b2a179f6..43cb8aaf57a2 100644
> --- a/arch/arc/Kconfig
> +++ b/arch/arc/Kconfig
> @@ -314,6 +314,10 @@ config ARC_HUGEPAGE_16M
>
> endchoice
>
> +config PGTABLE_LEVELS
> + int "Number of Page table levels"
> + default 2
> +
> config ARC_COMPACT_IRQ_LEVELS
> depends on ISA_ARCOMPACT
> bool "Setup Timer IRQ as high Priority"
> diff --git a/arch/arc/include/asm/page.h b/arch/arc/include/asm/page.h
> index 313e6f543d2d..df3cc154ae4a 100644
> --- a/arch/arc/include/asm/page.h
> +++ b/arch/arc/include/asm/page.h
> @@ -41,6 +41,17 @@ typedef struct {
> #define pgd_val(x) ((x).pgd)
> #define __pgd(x) ((pgd_t) { (x) })
>
> +#if CONFIG_PGTABLE_LEVELS > 2
> +
> +typedef struct {
> + unsigned long pmd;
> +} pmd_t;
> +
> +#define pmd_val(x) ((x).pmd)
> +#define __pmd(x) ((pmd_t) { (x) })
> +
> +#endif
> +
> typedef struct {
> #ifdef CONFIG_ARC_HAS_PAE40
> unsigned long long pte;
> diff --git a/arch/arc/include/asm/pgalloc.h b/arch/arc/include/asm/pgalloc.h
> index 0cf73431eb89..01c2d84418ed 100644
> --- a/arch/arc/include/asm/pgalloc.h
> +++ b/arch/arc/include/asm/pgalloc.h
> @@ -86,6 +86,28 @@ static inline void pgd_free(struct mm_struct *mm, pgd_t *pgd)
> }
>
>
> +#if CONFIG_PGTABLE_LEVELS > 2
> +
> +static inline void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmdp)
> +{
> + set_pud(pudp, __pud((unsigned long)pmdp));
> +}
> +
> +static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long addr)
> +{
> + return (pmd_t *)__get_free_page(
> + GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_ZERO);
> +}
> +
> +static inline void pmd_free(struct mm_struct *mm, pmd_t *pmd)
> +{
> + free_page((unsigned long)pmd);
> +}
> +
> +#define __pmd_free_tlb(tlb, pmd, addr) pmd_free((tlb)->mm, pmd)
> +
> +#endif
> +
> /*
> * With software-only page-tables, addr-split for traversal is tweakable and
> * that directly governs how big tables would be at each level.
> diff --git a/arch/arc/include/asm/pgtable-levels.h b/arch/arc/include/asm/pgtable-levels.h
> index 8ece75335bb5..1c2f022d4ad0 100644
> --- a/arch/arc/include/asm/pgtable-levels.h
> +++ b/arch/arc/include/asm/pgtable-levels.h
> @@ -10,6 +10,8 @@
> #ifndef _ASM_ARC_PGTABLE_LEVELS_H
> #define _ASM_ARC_PGTABLE_LEVELS_H
>
> +#if CONFIG_PGTABLE_LEVELS == 2
> +
> /*
> * 2 level paging setup for software walked MMUv3 (ARC700) and MMUv4 (HS)
> *
> @@ -37,16 +39,38 @@
> #define PGDIR_SHIFT 21
> #endif
>
> -#define PGDIR_SIZE BIT(PGDIR_SHIFT) /* vaddr span, not PDG sz */
> -#define PGDIR_MASK (~(PGDIR_SIZE - 1))
> +#else
> +
> +/*
> + * A default 3 level paging testing setup in software walked MMU
> + * MMUv4 (8K page): <4> : <7> : <8> : <13>
> + */
> +#define PGDIR_SHIFT 28
> +#if CONFIG_PGTABLE_LEVELS > 2
> +#define PMD_SHIFT 21
> +#endif
> +
> +#endif
>
> +#define PGDIR_SIZE BIT(PGDIR_SHIFT)
> +#define PGDIR_MASK (~(PGDIR_SIZE - 1))
> #define PTRS_PER_PGD BIT(32 - PGDIR_SHIFT)
>
> -#define PTRS_PER_PTE BIT(PGDIR_SHIFT - PAGE_SHIFT)
> +#if CONFIG_PGTABLE_LEVELS > 2
> +#define PMD_SIZE BIT(PMD_SHIFT)
> +#define PMD_MASK (~(PMD_SIZE - 1))
> +#define PTRS_PER_PMD BIT(PGDIR_SHIFT - PMD_SHIFT)
Maybe move these into the previous #if CONFIG_PGTABLE_LEVELS > 2?
> +#endif
> +
> +#define PTRS_PER_PTE BIT(PMD_SHIFT - PAGE_SHIFT)
>
> #ifndef __ASSEMBLY__
>
> +#if CONFIG_PGTABLE_LEVELS > 2
> +#include <asm-generic/pgtable-nopud.h>
> +#else
> #include <asm-generic/pgtable-nopmd.h>
> +#endif
>
> /*
> * 1st level paging: pgd
> @@ -57,9 +81,35 @@
> #define pgd_ERROR(e) \
> pr_crit("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
>
> +#if CONFIG_PGTABLE_LEVELS > 2
> +
> +/* In 3 level paging, pud_* macros work on pgd */
> +#define pud_none(x) (!pud_val(x))
> +#define pud_bad(x) ((pud_val(x) & ~PAGE_MASK))
> +#define pud_present(x) (pud_val(x))
> +#define pud_clear(xp) do { pud_val(*(xp)) = 0; } while (0)
> +#define pud_pgtable(pud) ((pmd_t *)(pud_val(pud) & PAGE_MASK))
> +#define pud_page(pud) virt_to_page(pud_pgtable(pud))
> +#define set_pud(pudp, pud) (*(pudp) = pud)
> +
> +/*
> + * 2nd level paging: pmd
> + */
> +#define pmd_ERROR(e) \
> + pr_crit("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
> +
> +#define pmd_pfn(pmd) ((pmd_val(pmd) & PMD_MASK) >> PAGE_SHIFT)
> +#define pfn_pmd(pfn,prot) __pmd(((pfn) << PAGE_SHIFT) | pgprot_val(prot))
> +#define mk_pmd(page,prot) pfn_pmd(page_to_pfn(page),prot)
> +
> +#endif
> +
> /*
> - * Due to the strange way generic pgtable level folding works, in a 2 level
> - * setup, pmd_val() returns pgd, so these pmd_* macros actually work on pgd
> + * Due to the strange way generic pgtable level folding works, the pmd_* macros
> + * - are valid even for 2 levels (which supposedly only has pgd - pte)
> + * - behave differently for 2 vs. 3
> + * In 2 level paging (pgd -> pte), pmd_* macros work on pgd
> + * In 3+ level paging (pgd -> pmd -> pte), pmd_* macros work on pmd
> */
> #define pmd_none(x) (!pmd_val(x))
> #define pmd_bad(x) ((pmd_val(x) & ~PAGE_MASK))
> @@ -70,6 +120,9 @@
> #define set_pmd(pmdp, pmd) (*(pmdp) = pmd)
> #define pmd_pgtable(pmd) ((pgtable_t) pmd_page_vaddr(pmd))
>
> +/*
> + * 3rd level paging: pte
> + */
> #define pte_ERROR(e) \
> pr_crit("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
>
> diff --git a/arch/arc/include/asm/processor.h b/arch/arc/include/asm/processor.h
> index e4031ecd3c8c..f28afcf5c6d1 100644
> --- a/arch/arc/include/asm/processor.h
> +++ b/arch/arc/include/asm/processor.h
> @@ -93,7 +93,7 @@ extern unsigned int get_wchan(struct task_struct *p);
> #define VMALLOC_START (PAGE_OFFSET - (CONFIG_ARC_KVADDR_SIZE << 20))
>
> /* 1 PGDIR_SIZE each for fixmap/pkmap, 2 PGDIR_SIZE gutter (see asm/highmem.h) */
> -#define VMALLOC_SIZE ((CONFIG_ARC_KVADDR_SIZE << 20) - PGDIR_SIZE * 4)
> +#define VMALLOC_SIZE ((CONFIG_ARC_KVADDR_SIZE << 20) - PMD_SIZE * 4)
>
> #define VMALLOC_END (VMALLOC_START + VMALLOC_SIZE)
>
> diff --git a/arch/arc/mm/fault.c b/arch/arc/mm/fault.c
> index 41f154320964..8da2f0ad8c69 100644
> --- a/arch/arc/mm/fault.c
> +++ b/arch/arc/mm/fault.c
> @@ -39,6 +39,8 @@ noinline static int handle_kernel_vaddr_fault(unsigned long address)
> if (!pgd_present(*pgd_k))
> goto bad_area;
>
> + set_pgd(pgd, *pgd_k);
> +
> p4d = p4d_offset(pgd, address);
> p4d_k = p4d_offset(pgd_k, address);
> if (!p4d_present(*p4d_k))
> @@ -49,6 +51,8 @@ noinline static int handle_kernel_vaddr_fault(unsigned long address)
> if (!pud_present(*pud_k))
> goto bad_area;
>
> + set_pud(pud, *pud_k);
> +
> pmd = pmd_offset(pud, address);
> pmd_k = pmd_offset(pud_k, address);
> if (!pmd_present(*pmd_k))
> diff --git a/arch/arc/mm/tlb.c b/arch/arc/mm/tlb.c
> index 34f16e0b41e6..77da83569b36 100644
> --- a/arch/arc/mm/tlb.c
> +++ b/arch/arc/mm/tlb.c
> @@ -658,8 +658,8 @@ char *arc_mmu_mumbojumbo(int cpu_id, char *buf, int len)
> IS_USED_CFG(CONFIG_TRANSPARENT_HUGEPAGE));
>
> n += scnprintf(buf + n, len - n,
> - "MMU [v%x]\t: %dk PAGE, %sJTLB %d (%dx%d), uDTLB %d, uITLB %d%s%s\n",
> - p_mmu->ver, p_mmu->pg_sz_k, super_pg,
> + "MMU [v%x]\t: %dk PAGE, %s, swalk %d lvl, JTLB %d (%dx%d), uDTLB %d, uITLB %d%s%s\n",
> + p_mmu->ver, p_mmu->pg_sz_k, super_pg, CONFIG_PGTABLE_LEVELS,
> p_mmu->sets * p_mmu->ways, p_mmu->sets, p_mmu->ways,
> p_mmu->u_dtlb, p_mmu->u_itlb,
> IS_AVAIL2(p_mmu->pae, ", PAE40 ", CONFIG_ARC_HAS_PAE40));
> diff --git a/arch/arc/mm/tlbex.S b/arch/arc/mm/tlbex.S
> index d08bd09a0afc..5f6bfdfda1be 100644
> --- a/arch/arc/mm/tlbex.S
> +++ b/arch/arc/mm/tlbex.S
> @@ -173,6 +173,15 @@ ex_saved_reg1:
> tst r3, r3
> bz do_slow_path_pf ; if no Page Table, do page fault
>
> +#if CONFIG_PGTABLE_LEVELS > 2
> + lsr r0, r2, PMD_SHIFT ; Bits for indexing into PMD
> + and r0, r0, (PTRS_PER_PMD - 1)
> + ld.as r1, [r3, r0] ; PMD entry
> + tst r1, r1
> + bz do_slow_path_pf
> + mov r3, r1
> +#endif
> +
> #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> and.f 0, r3, _PAGE_HW_SZ ; Is this Huge PMD (thp)
> add2.nz r1, r1, r0
> --
> 2.25.1
>
--
Sincerely yours,
Mike.
More information about the linux-snps-arc
mailing list